Time to Align! Modelling Musical Timelines for Music Information Retrieval and Digital Musicology

Aligning musical data – matching corresponding events across different representations of the same music – is central to Music Information Retrieval and Digital Musicology. Datasets span graphical representations (images, scans), logical encodings (scores, MIDI) and physical recordings (audio, video, sensor data), each imposing its own coordinate system for musical time. Integrating them requires costly alignment, but results are typically stored in ad hoc formats that are poorly documented and difficult to share and reuse. To address this, we present TimeToAlign!, a cross‑domain model for musical timelines and their alignments that organises time‑related data into continuous and discrete variants across three temporal domains – graphical, logical and physical – connected by typed conversion maps and explicit match claims. In addition to describing the model, we demonstrate it through five concrete use cases of increasing complexity: (1) aligning a historical piano roll with a digital score, (2) modelling flow control in a school song arrangement, (3) transferring annotations between graphical analyses, (4) propagating harmonic labels across hundreds of multimodal timelines of a Beethoven string quartet and (5) encoding the conceptual and temporal relationships in the genesis of a rock song. The model is implemented as timetoalign, an open‑source Python library with typed timelines, loaders for common formats (such as audio formats, MusicXML or CSV) and conversion maps. Runnable tutorial and how‑to notebooks, a documentation homepage and a collection of real‑world data from diverse repertoires accompany the library.